Closed dnwjddl closed 1 year ago
Hello. What do you mean by "not correct"? We used the weights from https://github.com/mapillary/inplace_abn . Are you able to use these weights?
Hi, Thank you for the comment :) I solved the problem that I previously asked about. Additionally, I tried to reproduce the code using the weights of ResNet101v1 and InPlace-ABN sync from that website, but the values are coming out lower than expected. I'm wondering if there might be any changed parameters or if I accidentally used different weights?
We used the weights from their repository, without changes. Can you detail what you are obtaining? What method are you reproducing? On which setting?
Hello, First of all, thanks for sharing such wonderful work!
I noticed that the pretrained weights currently available are not correct. Would it be possible to share the correct pretrained weights(
pretrained/resnet101_iabn_sync.pth.tar
)?Thank you!